Analyzing Stability of Convolutional Neural Networks in the Frequency Domain
This work addresses the stability issue in ConvNets for researchers and practitioners, but it is incremental as it builds on existing frequency domain methods without introducing a new paradigm.
The paper tackled the problem of analyzing and improving the stability of convolutional neural networks (ConvNets) against noise by using frequency domain analysis, revealing that trained ConvNets often pass most frequencies and are sensitive to low-magnitude noise, and showing that fine-tuning with noisy images can produce more stable networks.
Understanding the internal process of ConvNets is commonly done using visualization techniques. However, these techniques do not usually provide a tool for estimating the stability of a ConvNet against noise. In this paper, we show how to analyze a ConvNet in the frequency domain using a 4-dimensional visualization technique. Using the frequency domain analysis, we show the reason that a ConvNet might be sensitive to a very low magnitude additive noise. Our experiments on a few ConvNets trained on different datasets revealed that convolution kernels of a trained ConvNet usually pass most of the frequencies and they are not able to effectively eliminate the effect of high frequencies. Our next experiments shows that a convolution kernel which has a more concentrated frequency response could be more stable. Finally, we show that fine-tuning a ConvNet using a training set augmented with noisy images can produce more stable ConvNets.